MIT14 126S16 Globalgames
MIT14 126S16 Globalgames
1
Motivation
2
A partnership game
Invest Not-Invest
3
T is common knowledge
T<0
Invest Not-Invest
Invest TT
TT T
T
Not-Invest T
T
4
T is common knowledge
T>1
Invest NotInvest
Invest TT T
T
NotInvest T
T
5
T is common knowledge
6
T is common knowledge
Not-Invest Invest
Multiple
Equilibria
T
7
T is not common knowledge
8
Recall: Monotone supermodular games
G = (N,T,A,u,p)
T = T0 u T1 u … u Tn ( RM)
Ai compact sublattice of RK
ui : A u T
ui(a,.): T
ui(. ,t): A supermodular in ai,
has increasing differences in a and in (ai,t)
p(.|ti) is increasing function of ti—in the sense of 1st-order
stochastic dominance (e.g. p is affiliated).
Theorem: There exist BNE s* and s** such that
For each BNE s, s* t s t s**.
9
Conditional Beliefs given xi
T d xi – HKi
i.e. Pr(TTc_xi) = 1-F((xi-Tc)/H):=G(Tc_xi)
xj d xi + H Kj-Ki)
Pr(xjxjc_xi) = Pr(H(Kj – Ki) xjc – xi)
Pr(T Tcxj xjc_xi) =
1{T Tc} ((xjc T)/)( | ) decreasing in xi
because integrand decreasing in T and G(_xi)
FOSD G(_xic) whenever xi xic
cE[T|x ] = x
i i
10
Payoffs
11
Monotone BNE
Best response
Invest iff xi t Pr(sj = Not-Invest|xi)
Assume supp(T) = [a,b] where a < 0 < 1 < b.
xi < 0 si(xi) = Not Invest
xi > 1 si(xi) = Invest
A cutoff xi* s.t.
xi < xi* si(xi) = Not Invest; xi > xi* si(xi) = Invest
Symmetry: x1* = x2* = x*
x* = Pr(sj = Not-Invest|x*) = Pr(xj < x*|xi=x*) = 1/2
“Unique” BNE
12
Questions
13
T is not common knowledge
but the noise is very small
It is very likely that
Not-Invest
Invest
T
1/2
14
Risk-dominance
In a 2 x 2 symmetric game, a strategy is said to be
“risk dominant” iff it is a best reply when the other
player plays each strategy with equal probabilities.
Invest Not-Invest
15
Carlsson & van Damme
16
Risk Dominance
A B Suppose that (A,A) and
(B,B) are NE.
A u11,v11 u12,v12 (A,A) is risk dominant if
(u11-u21)(v11-v12)
B u21,v21 u22,v22 ¾ (u22-u12) (v22-v21)
17
Dominance, risk-dominance regions
Dominance region
Dia ={(u,v)| gia>0, gib<0}
Risk-dominance region
18
Model
19
Theorem
Suppose that
x is on a continuous curve C 4
(u(c),v(c)) Ra for each cC
(u(c),v(c)) Da for some cC.
Then A is the only rationalizable action at x
when H is small.
20
“Public” Information
T ~ N(y,W2) and HKi ~ N(0,V2)
Given xi,
T ~ N(rxi+(1-r)y, V2r)
xj ~ N(rxi+(1-r)y, V2(r+1))
r = W2/(V2+W2)
(Monotone supermodularity) monotone symmetric NE w/cutoff xc:
§ (1 r )(x c y ) ·
rx c (1 r )y Pr(x j d x c | x i xc ) )¨¨ ¸¸
© V r 1 ¹
Unique monotone NE (and rationalizable strategy) if
rx c (1 r )y Pr(x j d x c | x i xc )
is increasing in xc whenever zero, i.e.,
V < 2SW4(r+1)
21
Courtesy of Stephen Morris and Hyun Song Shin. Used with permission.
22
MIT OpenCourseWare
https://ocw.mit.edu
For information about citing these materials or our Terms of Use, visit: https://ocw.mit.edu/terms.